Dr. Yuxiang Zhang’s recent project release

UniTS: Unified Time Series Generative Model for Remote Sensing

Congratulations to Dr. Yuxiang Zhang and the research team on the release of their novel framework, UniTS. The project page is now live at https://yuxiangzhang-bit.github.io/UniTS-website/, with code available on GitHub (https://huggingface.co/datasets/YuxiangZhang-BIT/UniTS-Datasets-ckpt) and datasets accessible via Hugging Face (https://huggingface.co/datasets/YuxiangZhang-BIT/UniTS-Datasets-ckpt).

This study addresses a primary objective of satellite remote sensing: capturing the complex dynamics of the Earth's environment. While tasks such as reconstructing cloud-free images, detecting land cover changes, and forecasting surface evolution are critical, existing methods typically rely on specialized models tailored to single tasks, lacking a unified approach. UniTS (Unified Time Series Generative Model) introduces a general framework applicable to a wide range of tasks, including time series reconstruction, cloud removal, semantic change detection, and forecasting.

Based on the flow matching generative paradigm, UniTS constructs a deterministic evolution path from noise to targets under the guidance of task-specific conditions, achieving unified modeling of spatiotemporal features. Additionally, the team has constructed two high-quality multimodal time series datasets, TS-S12 and TS-S12CR, filling the gap for benchmark datasets in cloud removal and forecasting. Extensive experiments demonstrate that UniTS exhibits exceptional generative and cognitive capabilities across both low-level and high-level time series tasks.

The framework of UniTS applied to four remote sensing time series tasks.

Left: TS-S12 dataset; middle: Geographical distribution of ROIs; right: TS-S12CR dataset.

Project page: https://yuxiangzhang-bit.github.io/UniTS-website/

Github: https://github.com/YuxiangZhang-BIT/UniTS

Datasets: https://huggingface.co/datasets/YuxiangZhang-BIT/UniTS-Datasets-ckpt

Keywords: Satellite image time series, Time series reconstruction, Time series cloud removal, Time series semantic change detection, Time series forecasting, Flow matching